5.1 Framing an ML problem
You can tell whether a business question is even an ML question.
Before:03. Data Handling & AnalysisUnlocks:06. Deep Learning10. Production & MLOps13. Capstones, Portfolio & Interviews
Framing is the step before any code: what exactly is being predicted, for which unit, against which baseline — and whether machine learning is warranted at all. Most failed ML projects failed here, not in the modelling. It opens the module because everything downstream inherits the frame. The non-negotiable is the baseline: when predicting yesterday's value or one simple rule scores nearly as well as a model would, that is the answer, and it costs nothing to run.
Work through these
Supervised, unsupervised, self-supervised, reinforcement
The four broad settings differ in what supervision is available: labelled examples, none, labels derived from the data itself, or a reward signal. Placing a problem correctly narrows the method enormously.
Choosing the target and the unit of prediction
Deciding what exactly is being predicted, and for what unit, is a modelling decision made before any algorithm is chosen. Getting it wrong produces a technically correct model answering the wrong question.
Baselines you must beat
Before any model, establish what a trivial approach achieves, whether that is predicting the average or the most common class. A model that does not beat it is not a result.
When not to use ML at all
Some problems are better solved by a rule, a lookup, or fixing the process that generates the data. Recognising these saves months of work that would have produced a mediocre model.
Sign in to keep your progress.
Free resources
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.